Design of novel high-performance fuels with artificial intelligence: Case study for spark-ignition engine applications

The ever-increasing importance of both energy security and sustainability motivates the design of carbon-neutral petroleum replacements from renewable resources. Fuel candidates are conventionally selected from existing databases with limited scope. This work presents a novel artificial intelligence-based fuel design approach, which identifies molecules tailor-made for a particular application by screening millions of candidates. The approach is demonstrated by the design of fuel blending components for spark-ignition engines. A virtual pool consisting of 26.2 million fuel molecules is first developed by considering all possible combinations of predefined structural groups. The practical application potential of these molecules is evaluated based on the joint consideration of various properties estimated by artificial neural network-based quantitative structure–property relationship models. A two-stage design process is performed. In particular, a number of species with novel and complex structures are identified. These are expected to allow for high efficiency and low emissions simultaneously, but have not attracted previous investigation in the literature yet. • An AI-driven fuel design approach is proposed, enabling the identification of tailor-made fuel molecules by screening a vast virtual pool, overcoming the limitations of conventional methods that rely on existing databases with limited scope. • Machine learning-based QSPR models are employed to accurately predict fuel properties, ensuring a comprehensive evaluation of application potential for each candidate molecule. • A two-stage design process is demonstrated, showcasing the effectiveness of the approach in designing fuel blending components for sustainable petroleum replacements. • Complex and novel fuel molecules are identified, offering the potential to achieve both high efficiency and low emissions in spark-ignition engines.